AI & Content · April 17, 2026 · 9 min read
Can AI Write Blog Posts That Actually Rank? We Tested It
We published 200 AI-generated posts across 10 WordPress sites and tracked rankings for 6 months. The answer is yes — with specific caveats about topic selection, editing depth, and the posts that consistently underperformed.
By FluxWriter Team
The experiment
We ran a controlled test across 10 WordPress sites in 5 niches (personal finance, home improvement, pet care, travel, and SaaS tools). Each site received 20 AI-generated posts: 10 with light editing (grammar and fact-check only) and 10 with deep editing (rewriting sections, adding original examples, embedding personal experience).
All posts targeted keywords with search volume between 200–2,000/month and keyword difficulty below 30 (Ahrefs scale). None of the sites had significant existing domain authority.
Results at 6 months
Lightly edited posts:
- 34% ranked on page 1 for their primary keyword
- 51% ranked on pages 2–3
- 15% ranked beyond page 3 or received no impressions
Deeply edited posts:
- 61% ranked on page 1
- 31% ranked on pages 2–3
- 8% ranked beyond page 3 or received no impressions
The editing depth nearly doubled the page-1 rate. This was the most consistent finding across all five niches.
What ranked easily
Informational "how-to" and "what is" queries in low-competition niches. AI models are genuinely strong at covering a topic comprehensively and accurately when the subject is well-represented in training data. Queries like "how to clean a dryer vent" or "what is a Roth IRA conversion ladder" ranked well with minimal editing.
Comparison posts with structured tables. AI consistently produced clean comparison tables, which Google often features in rich results. These posts punched above their domain authority weight.
Long-tail question posts. "Why does my dog eat grass every morning" ranked within 3 weeks on a new pet site. The AI answered comprehensively, the brief included a FAQ section, and the post was 900 words.
What struggled to rank
Anything requiring genuine personal experience. "Best hiking boots for wide feet — a 6-month review" written by AI ranked significantly worse than equivalent posts from sites that had real testers. Google's product review guidelines are increasingly effective at detecting the difference.
Local SEO content. AI confidently generates city-specific content that often contains subtle errors — wrong neighborhoods, incorrect business hours, non-existent landmarks. These errors erode E-E-A-T signals and the posts performed poorly.
Highly competitive YMYL (Your Money, Your Life) topics. Medical, legal, and financial content in competitive niches requires established authoritativeness that AI content alone cannot signal. These posts need expert bylines, citations, and authority signals that go beyond the post content.
The editing protocol that produced the best results
Our deep editing protocol was:
- Run the draft through a fact-checker tool (Perplexity for claims, Google for proper nouns)
- Add one personal or first-hand example per major section
- Replace all AI-generated statistics with cited, primary-source statistics
- Rewrite the introduction from scratch
- Add a genuine conclusion with a specific recommendation
- Insert one contextually relevant internal link per 300 words
This took an average of 22 minutes per post. The page-1 ranking rate was 61% — a level that would require significantly more time from a human writing from scratch.
The verdict
AI can absolutely write blog posts that rank. The ceiling is lower for experience-dependent and YMYL content. The floor is higher than most people expect for informational content in low-to-medium competition niches.
The key variable is not the AI model — it's the brief quality, the editing investment, and the topic selection. Those three inputs are still human decisions.